Comparing Machine Learning and Deep Learning Approaches to Diagnose Epilepsy Disease

Abstract

Even though, epilepsy is recognized as a component of brain disease, and seizures in these diseases vary from person to person. This study aimed to implement some procedures based on machine learning and deep learning on EEG signal frequency, where they were compared with each other. Accordingly, to extract more accurate statistical features such as the Short-Time Fourier Transform (STFT) environment in terms of which artificial intelligent algorithms are the most effective applications for early diagnosis in epilepsy disease, is the significant point of the research. After removing the noise from the EEG signal, it entered the STFT environment, where its spectrogram was determined. In the STFT environment, the statistical features of the signal have been extracted, which are segmented at a frequency sampling rate of 173.61 Hz in the form of triangular windows. Finally, seven segmentations have been obtained, in each of which four features have been extracted, for a total of 28 statistical features. Technically, in terms of the deep learning algorithm, feature extraction is done by performing CNN models such as Xception, EfficientNetB0, ResNet50, and VGG16, which are given as inputs to a simple model with the label of being healthy or an epileptic patient, and then the classification of two classes has been applied. In addition, in the machine learning algorithm, basically, the EEG signals were transferred from the time domain to the STFT domain while they were placed next to each other, where we applied an ANOVA feature selection, and extracted 10 main features, which are given to the model as inputs. Furthermore, K-Fold Validation was performed in the dataset evaluation and division section, and eventually, the results were computed in the Support Vector Machine (SVM) model, which equaled 95%, Linear Discriminant Analysis (LDA) equaled 97.5%, and K-nearest neighbors (KNN) equaled 90%. As a conclusion, the study has been figured out that the deep learning algorithm has better performance than the machine learning algorithm in order to early diagnose epilepsy disease.